TL;DR
- Modular data centres are pre-engineered facilities built off-site as standardised blocks — ISO containers, skids, or pods — and delivered to site for rapid assembly.
- Time to operation is compressed from the 18-36 months of a stick-built facility to 6-12 months, sometimes less.
- Each module typically packages compute, cooling, power, and fire suppression into a sealed unit that connects to site utilities via standardised interfaces.
- Used for edge sites, rapid deployments, capacity expansions, sovereign in-country builds, and disaster-recovery sites.
Overview#
Modular data centres apply the principle of factory pre-fabrication to a domain that has historically been stick-built on site. Rather than designing a facility from scratch, sourcing parts, and assembling them in the field, modular operators use repeatable engineering: a standardised block (a container, a skid, or a building module) is fabricated in a controlled factory environment, shipped to site, and connected to utilities.
Benefits cascade from that. Engineering cost is amortised across many modules. Quality is more consistent. Lead times shrink because long-lead components are pre-purchased to a standard BOM. Capex is often lower per kW. The trade-off is reduced site-specific optimisation — a modular block is what it is; you accept its constraints rather than designing around your specific room.
Common Module Types#
| Type | Form | Typical capacity |
|---|---|---|
| IT module | ISO container or custom shell | 100-500 kW |
| Power module | Skid with UPS, switchgear, transformer | 0.5-2 MW per skid |
| Cooling module | Skid with chillers, pumps, CDU | 0.5-2 MW per skid |
| Mech-elec block | Pre-fab building section | 1-5 MW per block |
| All-in-one micro-DC | Single sealed unit | 20-100 kW |
| Multi-storey modular hall | Pre-fab building stacked | 5-30 MW per hall |
Trade-offs vs Stick-Built#
- Time to operation: 6-12 months for modular vs 18-36 months for stick-built — often the deciding factor in fast-moving AI markets.
- Capex: typically 10-20 % lower per kW than equivalent stick-built capacity at small/medium scale; the gap closes at hyperscale where economies of scale tip the other way.
- Customisation: modular constraints are rigid; if your workload needs an unusual cooling temperature, an unusual power topology, or unusual physical security, stick-built may be required.
- Density: modular cooling envelopes can match stick-built in modern designs — DLC-ready modules are commercially available.
- Site preparation: even modular needs ground, utilities, fibre, and planning consent. The factory build cannot eliminate site-side work.
- Reusability: containerised modules can be relocated if needs change; stick-built cannot.
Where Modular Wins#
- Edge and regional deployments where many similar sites need to be built quickly.
- Sovereign in-country builds where local construction expertise for hyperscale-class data centres is limited.
- Rapid capacity expansions to existing campuses — drop a module on the pad, plug into utilities.
- Disaster-recovery and surge capacity, including military and humanitarian operations.
- AI builds where time-to-revenue matters more than long-term optimisation.
Where Stick-Built Still Wins#
- Hyperscale campuses (50 MW+) where bespoke engineering pays back through scale.
- Sites with site-specific architectural, planning, or heritage constraints.
- Tier IV certification programmes where every system must be independently tested and signed off, sometimes requiring on-site witness testing that is harder for pre-built modules.
- Unusual environmental or seismic requirements that drive bespoke structural design.
Operational Pitfalls#
- Transport logistics: oversize loads, road permits, and site access need to be verified before fabrication starts.
- Site readiness: a module arriving before the pad, the utility hookup, or the fibre is ready creates expensive parking-lot storage.
- Vendor lock-in: each modular vendor has proprietary connectors, interfaces, and control systems. Mixing vendors in one site is harder than it sounds.
- Field commissioning: modules tested in the factory still need site-wide integration testing. Skipping this is a common source of go-live delays.
- Lifecycle: modules age as units; planning for refresh, lift-and-replace, or end-of-life disposal needs to be baked in from day one.